Infer logcdf of discrete transformations #7444
Conversation
ricardoV94
force-pushed
the
simplify_order_logp
branch
4 times, most recently
from
f042733
to
159da27
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #7444 +/- ##
==========================================
- Coverage 92.17% 92.17% -0.01%
==========================================
Files 103 103
Lines 17260 17217 -43
==========================================
- Hits 15910 15869 -41
+ Misses 1350 1348 -2
|
There was a problem hiding this comment.
Looks good as always @ricardoV94 :)
Just to clarify: when you say discrete transformations, do you mean monotonic transformations of discrete variables?
| backward_value = op.transform_elemwise.backward(value, *other_inputs) | ||
|
|
||
| # Fail if transformation is not injective | ||
| # A TensorVariable is returned in 1-to-1 inversions, and a tuple in 1-to-many | ||
| if isinstance(backward_value, tuple): | ||
| raise NotImplementedError | ||
|
|
||
| is_discrete = measurable_input.type.dtype.startswith("int") |
There was a problem hiding this comment.
Shall we use discrete_types here?
Line 32 in 43dc4be
| return max_rv | ||
| measurable_max_class = ( | ||
| MeasurableMaxDiscrete if latent_base_var.type.dtype.startswith("int") else MeasurableMax | ||
| ) |
Yes. Right now only add and neg transforms are supported. We could add integer mul as well. Not keen to add support for integer->reals transforms just yet. Otherwise for continuous RVs we have a single case of non monotonicity we support (pow with negative exponent): pymc/pymc/logprob/transforms.py Line 260 in 43dc4be |
ricardoV94
force-pushed
the
simplify_order_logp
branch
from
159da27
to
e5899b4
Compare
Description
Allow logcdf inference for discrete transformations
With this we can also remove special case for MaxNeg (used to derive logprob of Min operations)
📚 Documentation preview 📚: https://pymc--7444.org.readthedocs.build/en/7444/